Table 3. The statistical comparison of synthesized results from slices involved in Figure 9. PIR and SIR of U-net, c-GAN, and the proposed bi-c-GAN were presented. Overall achievements acted as references.
| Models | Slice 1 | Slice 2 | Slice 3 | Slice 4 | Overall |
|---|---|---|---|---|---|
| U-net | |||||
| PIR (%) | 59.8±19.9 | 60.0±18.6 | 46.7±27.0 | 48.0±32.1 | 53.5±25.8 |
| SIR (%) | 29.8±19.6 | 25.8±20.2 | 21.3±17.7 | 27.4±23.0 | 26.1±20.5 |
| c-GAN | |||||
| PIR (%) | 58.2 ±14.7 | 57.1±12.3 | 50.6±24.8 | 51.0±30.7 | 54.1±22.3 |
| SIR (%) | 29.5±18.7 | 25.3±19.2 | 21.7±17.2 | 27.7±22.7 | 26.0±19.9 |
| Bi-c-GAN | |||||
| PIR (%) | 64.9±19.4 | 67.8±23.8* | 53.5±25.7 | 57.3±30.0 | 60.8±25.7 |
| SIR (%) | 30.2±20.0* | 26.5±21.2 | 21.9±17.6 | 28.3±22.8 | 26.7±20.7 |
Values are presented in the format of mean ± stand deviation. *, the highest improvement ratio. c-GAN, conditional generative adversarial network; Bi-c-GAN, bi-task c-GAN.